Full Transcript
Carl Lewis:
Welcome to the Connected Enterprise Podcast. I’m Carl Lewis, your host from Vision33, and my guest is Emrah Gultekin from Chooch AI. It's a pleasure to have you with us today, Emrah.
Emrah Gultekin:
Thank you, Carl. I enjoy being here.
Carl Lewis:
Please tell us about your background, Chooch AI, and your role there.
Emrah Gultekin:
We're a visual AI company based in Silicon Valley, and we basically clone human visual intelligence into machines. It sounds fancy, but it’s just teaching machines how to tag things as humans would in their specific expertise. My background is in business, and I’ve developed many companies; this is my seventh. Three succeeded, three went sideways, and one failed. As good entrepreneurs do, we try to fail fast. We started developing this type of AI four years ago, and we've been fortunate to be part of a good ecosystem for developing and deploying these AI systems. We're also fortunate to be integrating with large enterprises as we move forward.
Carl Lewis:
You and I have one thing in common: I tried to start a couple of companies. But I discovered it wasn't in my DNA because I couldn't sleep at night when I was the owner. I quickly figured out how to get a good job and parlayed myself into this role.
Emrah Gultekin:
Sounds familiar.
Carl Lewis:
Visual is a unique and creative space. What trends around automation are people talking about and doing in this industry? Many people talk about it, but you have clients who are truly doing it. At conferences, all you hear is “AI, machine learning, AI, machine learning,” but they don't have any live examples. So, who's really doing it, how well are they doing it, and how important is it to their business? Tell us that kind of stuff.
Emrah Gultekin:
That's a good point. People talk about this a lot. It's conceptualized, but we don't see it much in deployment, and it’s not happening as quickly as we want. One reason is that deep learning is part of machine learning, which is part of AI – but deep learning frameworks didn’t exist to the public until 2014 or 2015. These are TensorFlow (which is Google’s), Gluon, MXNet, and PyTorch (which is Facebook's). And although they exist today, and many developers play around with them and try to understand how they work, the components are unstable. It's difficult to 1) get them to work, 2) get them to work at scale, and 3) deploy them. And besides deep learning frameworks still being in development, you also have to build neural nets around them.
Emrah Gultekin:
Another reason is the processors. The theories and math for AI have been around conceptually for 40 years, and there haven’t been any big scientific breakthroughs in 10 or 15 years. But the processing power wasn't there. Now, with the processing power – the GPUs (specifically Nvidia's GPUs) – you can make these predictions. You may do the AI training and inferencing quickly on them. That’s helped in development, but you still have engineering risk. And product risk. Those are the issues we have in AI today. You'll hear about it because they work in very narrow settings – which is another thing we've been trying to solve: how do we generalize the AI? How do we make it generic so people can use it in different situations?
Emrah Gultekin:
We've been doing it on the cloud, and recently we've deployed onto the edge, which means using visual AI on the edge, in disconnected environments, and doing the inferencing there. Some trends have been that. Until you get the technology right, you won’t get mass consumption of these products. If the technology doesn't work correctly, in scale, you’ll have numerous issues.
Carl Lewis:
That's interesting. I like to say that if you're involved at this stage, you're building the airplane in the air. Is that right?
Emrah Gultekin:
Yes, and it's scary because we look for resources, and there’s nothing. You have to keep innovating yourself, and that’s an issue in AI. We're all in the same boat.
Carl Lewis:
There's a lot of personal investment that goes into this stage.
Emrah Gultekin:
Yes, and people get the idea that AI is here. And AI is here – at least, theoretically. Practically, we're maybe 10 years out. We don’t have self-driving cars and stuff like that, although companies are putting billions into it. And the problems I mentioned are why we don’t have them. They’re the problems everyone is trying to solve. And they'll be solved. But it takes a lot of resources.
Carl Lewis:
I hear you. Working ‘most of the time’ is unacceptable for a car. Are there other challenges to deploying this technology, even in that narrow window you mentioned?
Emrah Gultekin:
The challenges in deployment are different from the challenges in technology. Deployment challenges include getting companies to understand how AI benefits them. We struggle with this. We tell people it will save them millions. One client is saving hundreds of millions of dollars a year using our AI. Another one can deploy to 20,000 locations instantly. But the problem is, it does disrupt their internal business and workflows until it’s implemented. People want change, but they don't want to change everything simultaneously. There's a lot of handholding and training. We have to implement it into their workflows. It’s like the internet in the early '90s. Clients said, “Why should I have a website?” We said, “It’s the future.” They said, “Prove it.” But it's hard to prove because it hasn't happened yet. Then smart guys build ecommerce sites out of their garages and take over.
Emrah Gultekin:
Disrupting current enterprise practices is an issue. Not disrupting the market, per se, but disrupting what clients are doing in their workflows. They think it's hard, and it is. It's hard to implement these things. But we're trying to make it easier for our clients to adopt AI.
Carl Lewis:
You're working with early adopters, right?
Emrah Gultekin:
Early adopters and people who want to get ahead. They think, “If I don't do this, my competitors will. And then we’re done because they’ll leapfrog over everybody.” So reluctantly, yes, people want to do this. But they also don’t want to leave their comfort zones.
Carl Lewis:
You said maybe in 10 years we'll be there, but that journey will be fraught with new things that add to the technology that makes AI tick. What’s the next big thing?
Emrah Gultekin:
This COVID-19 crisis has accelerated the need for human and machine interaction. The classical thing is human to human, but soon it will be human to machine. The machines must be smarter at what they're doing, so you’ll hear more AI success stories. Whether it's in the medical, security and safety, or geospatial fields, there are already success stories. And more are coming. We recently read something for UCSD, for example. They're doing radiology. It's hard to predict which fields will flourish from this, but there’s healthcare. A lot of money is going into self-driving cars. Automation, manufacturing, safety, OSHA compliance, etc. It's hard to predict, but we’ll see many AI applications and success stories.
Carl Lewis:
It feels like it.
Emrah Gultekin:
Our healthcare company will save hundreds of millions a year using our visual AI system.
Carl Lewis:
It seems like, given the stress on our hospital systems and medical suppliers during this time, there should be a lot of innovation in that industry, so they're ready next time.
Emrah Gultekin:
That's an important point. We think of healthcare as a single industry, but we see healthcare going to every industry. Some clients have facial authentication, but now they want to check temperatures or if masks are on. Or detect coughing. Healthcare is spreading its wings over all verticals as we move out of the COVID crisis.
Carl Lewis:
I’m going to switch directions to more personal stuff. You've started seven businesses, so you've been doing this a while, although you look like a young guy compared to me. Communication changes a lot. I can only imagine how it’s been affected by everyone working from home. Not to mention the stress the networks are under! When I started in business, we did a lot by fax. My first telephone was in a suitcase; then I had a giant thing in the car. Then it was a brick I hauled around. Then it was three devices to do everything one smartphone can do today. And don't forget email – that was a life-changer. How do you do most of your business communication, and is it changing?
Emrah Gultekin:
Great point, Carl. We use 14 tools for communication and enterprise. When we discovered that, we were like, “What’s going on?” Because 10 years ago, we were using one. Now we're using 14 – everything from accounting to marketing to sales to regular communication between teams. This increase in tools sometimes gets confusing. We're doing it to be more efficient and faster. We have email, but also Zoom for calls and Slack for project management. We use Trello and Asana. We see more and more tools, but I predict we’ll see a consolidation where one company comes in and sits on top of these. That's happened in the past.
Emrah Gultekin:
What we're doing in visual AI is part of this. We're visual beings, and since 83% of what we do is visual, more efficient, faster visual AI will be critical.
Carl Lewis:
When I heard people hacked into the educational Zoom stuff they're doing with kids, I thought some sort of visual AI that said, “Hey, he doesn't belong here” would be a real help.
Emrah Gultekin:
Exactly. Making sure the right people are there and that anyone who shouldn’t be there is kicked out automatically by AI. We do this for ecommerce clients – make sure their uploaded images or videos are the correct content. You can't manually do it. It's millions of streams.
Carl Lewis:
Too much, yes.
Emrah Gultekin:
It's humanly impossible, so you need AI to control it. That's also part of our mission.
Carl Lewis:
Absolutely. You said one frustration people have is that there are so many tools. Like for this podcast, we're using GoToMeeting because I've had good results with it. But we also use Microsoft Teams, and other people use Zoom. It would be great if there were a consolidator to make it so that no matter which tool you use, you’re all together when you connect with other people.
Emrah Gultekin:
Exactly. I think we’ll see that happen in the next few years.
Carl Lewis:
I like the concept, that's for sure. In your business, you serve as a third party, helping companies reach their goals and implement solutions. What do you think, for both your customers and you, are the biggest challenges when working with a third party?
Emrah Gultekin:
We're trying to make things seamless. And consultants, although they’re great, bill by the hour. So, there's a reverse incentive to make things more difficult or take longer. We're trying to do the opposite. We want to make things easier so we can scale to everybody. You can't scale a consulting business. Integration is the third-party challenge across all verticals. People don’t have developers on site to integrate these tools, so you must create things where users can sign into a dashboard and use them out of the box. That's what we've focused on – no-code implementation or visual AI. And the engineering team says, "That's impossible." But you make it possible because if you don't, you can’t scale. So third-party integration, that's a huge issue.
Emrah Gultekin:
Managing expectations is another thing. When we say “AI,” we’re not talking about the Terminator – we're talking about a computational tool. It's not a robot like in the movies. So, when we’re implementing an AI tool, we must manage people’s expectations, and it’s a huge task. We believe in science fiction, but we also know the limitations of current technology. It's important to address that up front and understand the client’s workflow and where AI fits to make their lives easier. I also think third parties should help clients sell the technology internally because if they can’t sell it to their employees, you can’t onboard it. Sometimes that’s offering tools like dashboards or things that work out of the box so they can demonstrate it to their peers. That's what we've done. We've made it remarkably simple.
Carl Lewis:
That's a good point. Some people in consulting try not to have a billable concept, but it's difficult to get away from. If you've done that, congratulations.
Emrah Gultekin:
We do have a billable concept, but we try to get away from it. We do it because if something will take 100 hours to implement, there needs to be compensation. You must be able to cover the costs. But we're trying to get 100 hours down to 10 hours down to five hours down to five minutes.
Carl Lewis:
Have you automated parts of your business?
Emrah Gultekin:
Yes. There are two main components of AI – training and inferencing. After an AI is trained, you must be able to inference. And we've automated both, which took many years. Then you have the dataset collection part, which is pre-training. Dataset collection, annotation, and labeling must be done for you to service it to the training engine. The training engine then understands what you've done on the dataset and puts it into the inference engine. We've automated the training, the inference engine, and the dataset annotation. Are we where we want to be? No. There's always room for improvement, and we're trying to make it even more seamless. That's our mission – to make it seamless.
Carl Lewis:
You just made me think of something. When I go to a grocery store with a list of 10 vegetables, I’d like it if I didn't have to enter the vegetable’s name or number in the self-checkout. I want to set it on the scale and have the machine say, "Those are bananas." It would save me so much time. I don't type well, and then they give me a screen in a different position, and it's not convenient. That'd be right up your ally.
Emrah Gultekin:
It'll come. We were thinking the same thing the other day. I’ve bought bananas, and I saw bananas on the screen, and I thought, “Did they implement this already?” No – my wife had already typed it in! But I was like, “Damn, this could be done.” That's a great example.
Carl Lewis:
And the dataset would be every fruit and vegetable in the fresh section of a grocery store.
Emrah Gultekin:
Exactly. For the datasets, you say, "This is a banana."
Emrah Gultekin:
And the AI says, "This is a banana. I'll remember that." Then, the next time it sees a banana, it says, "This is a banana." But you can get into more detail. Like this is a type of fracture on the tibia. That's why we say we're cloning human visual expertise. Anything you're an expert in, you can teach the AI, and that's part of the datasets.
Carl Lewis:
Excellent. When you deploy things at businesses, do they measure how well it's working? Do they have KPIs around that?
Emrah Gultekin:
That's an important point. Since we're automating human intelligence, the only KPIs are, “How long would it take a human to do this?” “How many humans would I need to do this?” “How well is AI performing this?” “How does it save me money, and how does it make me money?” And we're seeing enormous amounts of improvement, like 80 to 100 times better, in their processes. And they can push things out to their clients faster. But there are things AI can't do, so you still need humans for a few reasons. One is to teach the AI; the other is to check the AI. You predicted something, but is it correct? You must ensure the models are working, so you still have humans checking things – just much less than before.
Carl Lewis:
You’ve given us great information today. It's good to talk to someone who's really doing it. It brings it down to earth for those of us who just hear the buzzwords. It was also great to meet someone who spent time in Turkey like I did as a boy. Thank you for reminiscing with me. And I have a favor to ask: if one of these big things comes about that you think my listeners and I should know about, contact me so we can do this again. You’re a great source of information.
Emrah Gultekin:
Absolutely, Carl. Thank you.
Carl Lewis:
I try to finish these in 20 to 30 minutes, so when we get back to driving, people can listen during their commute. Again, thank you. And everybody out there, until we hear from each other again, stay connected.
